[WSJ] Samsung, Nvidia Shy Away from Server Chip Battle

[AtenRa]

http://blogs.wsj.com/digits/2014/06/24/samsung-nvidia-shy-away-from-server-chip-battle/?mod=ST1

Didnt expect that, is it the small market share of ARM Servers or they dont want to engage to low margins markets ??

But Samsung recently decided to drop the server chip effort and discontinue work being carried out by a team in Milpitas, Calif., a person familiar with the situation said.
That’s not far from a longtime outpost for Samsung’s chip unit in San Jose, where the company is building a new research and development center.
“In response to recent changes in the business environment and the emergence of new R&D opportunities, Samsung Semiconductor has refocused some of its R&D efforts,” a spokeswoman for the company said. “The scope of the new campus project will not be affected.”


Another company leaving the server-chip push to others, at least for the moment, is Nvidia.
An Nvidia spokesman stressed that Project Denver will still be a high-performance chip, but will initially be part of the Tegra line focusing on mobile devices. He said the company is keeping its options open about pursuing servers at some point in the future.

 

[Homeles]

I didn't even know Samsung was gunning for the server market.

 

[CHADBOGA]

Intel & x86 have destroyed all the other RISC Server efforts, with the exception of POWER(which is just clinging on), what magic pixie dust does ARM have, that would allow companies to do better than all those from the recent past, who fell away with their RISC designs?

 

[AtenRa]

Intel & x86 have destroyed all the other RISC Server efforts, with the exception of POWER(which is just clinging on), what magic pixie dust does ARM have, that would allow companies to do better than all those from the recent past, who fell away with their RISC designs?

For specific segments ,

1. Cheaper : lower ownership cost
2. Lower power : lower operating cost
3. Lower TDP : lower operating cost
4. More cores per rack : lower operating cost

 

[ShintaiDK]

For specific segments ,

1. Cheaper : lower ownership cost
2. Lower power : lower operating cost
3. Lower TDP : lower operating cost
4. More cores per rack : lower operating cost

If that was only true.

Atom wins the metrics there, assuming you are in for the 1-2% that want microservers to start with.

Virtualization isnt making it any easier for miceservers either.

 

[ShintaiDK]

Didnt expect that, is it the small market share of ARM Servers or they dont want to engage to low margins markets ??

Its not just low margins, its also tiny volume. And neither of the companies got any integration fabric.

 

[NTMBK]

Intel & x86 have destroyed all the other RISC Server efforts, with the exception of POWER(which is just clinging on), what magic pixie dust does ARM have, that would allow companies to do better than all those from the recent past, who fell away with their RISC designs?

So you mean that the underperforming, consumer oriented architecture, backed by massive growth in an expanding market, destroyed the high margin, entrenched, "real man performance" incumbents? And don't forget that it was AMD, not Intel, which really cracked the server market wide open. Intel wanted to go for Itanium...

 

[Arachnotronic]

So you mean that the underperforming, consumer oriented architecture, backed by massive growth in an expanding market, destroyed the high margin, entrenched, "real man performance" incumbents? And don't forget that it was AMD, not Intel, which really cracked the server market wide open. Intel wanted to go for Itanium...

Backed by massive growth, huge revenues, *and* leading edge fabs pumping out tons of high margin chips, driving scale.

ARM is an IP licensing firm, so for ARM to succeed in the datacenter, its licensees need to wage war on Intel, not ARM itself. Who's gonna be the one to do it? AMD which has perpetually failed to compete with Intel in servers since 2008 and spends in a year on R&D company wide less than what Intel's data-center group alone spends in addition to the synergies that Intel gets from the rest of its efforts? Cavium Networks whose revenue base is smaller than what Intel gives away to its shareholders in dividends every 3 months? Applied Micro which just barely tripped over itself to get a 40nm chip out the door after hyping it up to no end?

And, funny thing is, none of the competitors that will be fighting Intel have the revenues/scale to fight Intel profitably in this market. If even Samsung didn't want anything to do with this (remember: Samsung has fabs and money), why should anybody else think it's a good idea?

ARM servers are basically a great way for these relatively small, unknown networking chip companies to ride the ARM hype train. Eventually economic reality will settle in and these players will realize that in order to survive, focusing R&D on areas that Intel isn't trying to take over is a much better plan than trying to take Intel head-on.

 

[NTMBK]

Backed by massive growth, huge revenues, *and* leading edge fabs pumping out tons of high margin chips, driving scale.

But Intel's "massive growth" phase is over- the PC market is stagnant, and they are failing in the phone and tablet markets. Where will the new volume come from to fill those leading edge fabs?

 

[Vesku]

Seems to me Nvidia wants to sell 'shovels' to the ARM server 'prospector' companies. Samsung might want to do the same or may just be wary of trying to carve out a niche from Intel given Intel's dominance of the server market.

 

[jj109]

So you mean that the underperforming, consumer oriented architecture, backed by massive growth in an expanding market, destroyed the high margin, entrenched, "real man performance" incumbents? And don't forget that it was AMD, not Intel, which really cracked the server market wide open. Intel wanted to go for Itanium...

Between these two graphs, I'm having trouble seeing where AMD did any cracking at all.

 

[NTMBK]

Between these two graphs, I'm having trouble seeing where AMD did any cracking at all.

They introduced AMD64- 64 bit both opened the high end, and allowed the dominance of the midrange to continue when 32 bit wouldn't cut it any more. Though you are correct, I had underestimated just how much volume had been taken by x86 before the 64 bit changeover

 

[ShintaiDK]

But Intel's "massive growth" phase is over- the PC market is stagnant, and they are failing in the phone and tablet markets. Where will the new volume come from to fill those leading edge fabs?

Intel is growing rapidly in tablets.

And in terms of stagnant PC market. You may have missed this:
http://newsroom.intel.com/community...ll-year-revenue-and-gross-margin-expectations

x86 is back in growth.

 

[ShintaiDK]

They introduced AMD64- 64 bit both opened the high end, and allowed the dominance of the midrange to continue when 32 bit wouldn't cut it any more. Though you are correct, I had underestimated just how much volume had been taken by x86 before the 64 bit changeover

It took almost 7 years from AMD64 was launched to actually becoming something that could be widely used in the server space. Until then 32bit was still the king for something like 95%.

The selling point wasnt AMD64, but the interconnect and core performance.

 

[teejee]

Intel is growing rapidly in tablets.

And in terms of stagnant PC market. You may have missed this:
http://newsroom.intel.com/community...ll-year-revenue-and-gross-margin-expectations

x86 is back in growth.

Intels profit is not the same as growing PC market. Not even close. Especially not since it is close to monopoly due to underperforming AMD.
Independent analysts predict no PC growth the next 4 years.
http://www.theguardian.com/technology/2014/jun/13/pc-market-doldrums-will-continue-to-2018-says-idc
In my opinion x86 is Intels biggest threat, it is so easy for Intel to make profit short term from x86 that Intel won't react in time for major changes like the move to ARM ISA.

 

[Idontcare]

http://blogs.wsj.com/digits/2014/06/24/samsung-nvidia-shy-away-from-server-chip-battle/?mod=ST1

Didnt expect that, is it the small market share of ARM Servers or they dont want to engage to low margins markets ??

I suspect that this is a case where quite simply the projected global TAM for this market segment over the next 5-8yrs results in a number that in no way justifies the R&D expenditures necessary to create the niche market in the first place.

Niche markets tend to require organic growth opportunities for long term viability to be possible.

If you have to dump in $5B to jump-start a market segment that at best is projected to max out at $2B per year revenue with mid-30's margins then that is a tough sell to a board of directors who are beholden to shareholders.

Pre-IPO startups (like Calxeda) don't have shareholders or analysts to worry about justifying their decisions to, they merely need to dupe a handful of venture capitalist managers into funding their pie-in-the-sky powerpoint schemes dreams a year at a time while hoping that someone like Samsung or Nvidia (or Facebook) will get scared they will be late to the emerging market party and buy them out for a few billion.

 

[meloz]

http://blogs.wsj.com/digits/2014/06/24/samsung-nvidia-shy-away-from-server-chip-battle/?mod=ST1

Didnt expect that, is it the small market share of ARM Servers or they dont want to engage to low margins markets ??

For servers it is important to make sure your hardware actually, you know, works with the software. That an update of the linux kernel won't render a box useless. These ARM clowns know all about selling disposable junk to consumers, but they cannot pull the same stunt in the server segment. "Just buy a new model" works for mobiles and tablets with Crapdroid, not so much in the data center.

 

[jpiniero]

Intel's server growth is mostly coming from the 'cloud companies' (eg: Google). Those companies seem more interested in various methods of custom functions to accelerate their processes than a general purpose processor. If they wanted to do it with ARM they could just do it themselves.

 

[Idontcare]

Intel's server growth is mostly coming from the 'cloud companies' (eg: Google). Those companies seem more interested in various methods of custom functions to accelerate their processes than a general purpose processor. If they wanted to do it with ARM they could just do it themselves.

While I have no doubt that "some" of Intel's server growth is coming from 'cloud companies', how do we know that "most" of the server growth is coming from that segment?

Does Intel do revenue breakouts by business division? (honest question, I wasn't aware Intel went into that granularity with their revenue streams) Is this a new thing with BK becoming CEO?

 

[ShintaiDK]

I would like to see some numbers too. Revenue breakup of the different server segments.

 

[witeken]

But Intel's "massive growth" phase is over- the PC market is stagnant, and they are failing in the phone and tablet markets. Where will the new volume come from to fill those leading edge fabs?

Smartphones and tablets.

I forgot internet of things.

 

[mrmt]

While I have no doubt that "some" of Intel's server growth is coming from 'cloud companies', how do we know that "most" of the server growth is coming from that segment?

Does Intel do revenue breakouts by business division? (honest question, I wasn't aware Intel went into that granularity with their revenue streams) Is this a new thing with BK becoming CEO?

They don't, they only report a single number for their operating semgment, and a few comments in the Q&A and SEC fillings, like this one:

http://www.intc.com/secfiling.cfm?filingID=50863-14-32

Net revenue for the DCG [Data Center Group] operating segment increased by $310 million, or 11%, in Q1 2014 compared to Q1 2013. DCG platform average selling prices and unit sales were up 8% and 3%, respectively. Our server platform unit sales continued to benefit from growth in Internet cloud computing and networking with continued strengthening of the enterprise market segments.

From what they say I don't think the growth comes only from cloud computing demand.

 

[witeken]

They don't, they only report a single number for their operating semgment, and a few comments in the Q&A and SEC fillings, like this one:



From what they say I don't think the growth comes only from cloud computing demand.

Everything is here: Annual Report 2013

 

[Tristor]

Intel's server growth is mostly coming from the 'cloud companies' (eg: Google). Those companies seem more interested in various methods of custom functions to accelerate their processes than a general purpose processor. If they wanted to do it with ARM they could just do it themselves.

That's why announcements like this are really interesting: http://www.extremetech.com/extreme/...h-integrated-fpga-touts-20x-performance-boost

It provides an opportunity for existing x86-centric workloads to be accelerated through custom logic in the FPGA portion. Can you imagine something like OpenStack being optimized heavily for it's most compute intensive operations in that FPGA, or having compute intensive common workloads being optimized and having that optimization exposed to tenant systems above?

Huge opportunities.

 

[Exophase]

That's why announcements like this are really interesting: http://www.extremetech.com/extreme/...h-integrated-fpga-touts-20x-performance-boost

It provides an opportunity for existing x86-centric workloads to be accelerated through custom logic in the FPGA portion. Can you imagine something like OpenStack being optimized heavily for it's most compute intensive operations in that FPGA, or having compute intensive common workloads being optimized and having that optimization exposed to tenant systems above?

Huge opportunities.

I can imagine the biggest players using this product as a stepping stone to develop their own SoCs with hard logic that implements the same functions more efficiently than an FPGA